dev-report 0.9.3

What it does

dev-report defines the report format every other crate in the dev-* suite emits. AI agents need machine-readable evidence of what passed, what failed, and why. This crate provides that schema.

Why it exists

A test runner that prints colored checkmarks to a TTY is unreadable to an AI agent. The agent needs:

• A stable, versioned schema
• Verdicts separated from logs
• Enough evidence to decide accept / reject / retry / escalate

dev-report is that schema.

Quick start

Add to Cargo.toml:

[dependencies]
dev-report = "0.9.3"

Opt-in features:

[dependencies]
dev-report = { version = "0.9.3", features = ["terminal", "markdown"] }

Build a report:

use dev_report::{Report, Verdict, Severity, CheckResult};

let mut report = Report::new("my-crate", "0.1.0")
    .with_producer("my-harness");

report.push(CheckResult::pass("compile"));
report.push(CheckResult::pass("test::unit").with_duration_ms(42));
report.push(
    CheckResult::fail("test::round_trip", Severity::Error)
        .with_detail("expected 42, got 41")
);

report.finish();

let verdict = report.overall_verdict();   // Verdict::Fail
let json = report.to_json().unwrap();     // ready to write to disk or stdout

Tags and evidence

Tags filter checks by category. Evidence attaches typed, decision-grade data without forcing the consumer to parse the free-form detail string.

use dev_report::{CheckResult, Evidence, Report};

let mut report = Report::new("my-crate", "0.2.0");

report.push(
    CheckResult::pass("bench::parse")
        .with_tag("bench")
        .with_duration_ms(7)
        .with_evidence(Evidence::numeric("mean_ns", 1234.5))
        .with_evidence(Evidence::numeric("baseline_ns", 1100.0))
        .with_evidence(Evidence::kv(
            "env",
            [("CI", "true"), ("RUST_LOG", "debug")],
        ))
        .with_evidence(Evidence::file_ref_lines("site", "src/parse.rs", 10, 20)),
);

// Filter by tag without parsing names.
let bench_checks: Vec<_> = report.checks_with_tag("bench").collect();
assert_eq!(bench_checks.len(), 1);

The four evidence kinds are:

Both tags and evidence are additive: v0.1.0 reports deserialize as v0.9.x reports with empty collections, and reports with no tags or evidence omit those keys from the JSON output.

Diffing two reports

Compare a current run against a baseline to flag regressions:

use dev_report::{CheckResult, DiffOptions, Report, Severity};

let mut baseline = Report::new("crate", "0.1.0");
baseline.push(CheckResult::pass("hot_path").with_duration_ms(100));

let mut current = Report::new("crate", "0.1.0");
current.push(CheckResult::fail("hot_path", Severity::Error).with_duration_ms(200));
current.push(CheckResult::pass("new_check"));

let diff = current.diff_with(
    &baseline,
    &DiffOptions {
        duration_regression_pct: Some(20.0),
        duration_regression_abs_ms: None,
    },
);

assert_eq!(diff.newly_failing, vec!["hot_path".to_string()]);
assert_eq!(diff.added, vec!["new_check".to_string()]);
assert!(!diff.is_clean());

Diff exposes newly_failing, newly_passing, severity_changes, duration_regressions, added, removed. All vectors are sorted by check name so two diffs of the same input pair are byte-equal.

Aggregating multiple producers

A single CI run usually invokes several producers (dev-bench, dev-fixtures, dev-async, ...). MultiReport aggregates them while preserving each check's (producer, name) identity:

use dev_report::{CheckResult, MultiReport, Report, Severity};

let mut bench = Report::new("crate", "0.1.0").with_producer("dev-bench");
bench.push(CheckResult::pass("hot_path"));

let mut chaos = Report::new("crate", "0.1.0").with_producer("dev-chaos");
chaos.push(CheckResult::fail("recover", Severity::Critical));

let mut multi = MultiReport::new("crate", "0.1.0");
multi.push(bench);
multi.push(chaos);
multi.finish();

let json = multi.to_json().unwrap();

Output formats

Four opt-in features render a Report (and MultiReport) into other formats. All are pure functions; JSON remains the only round-trippable wire format.

• terminal — to_terminal / to_terminal_color (ANSI). 80-column friendly. No new dependencies.
• markdown — to_markdown emits a CommonMark-compatible document preserving every fact (verdict, severity, tags, evidence, durations). No new dependencies.
• sarif — to_sarif emits a SARIF 2.1.0 document. Only Fail and Warn checks are included (SARIF is a defect-report format). Severity maps to SARIF level: Critical/Error → error, Warning → warning, Info → note. Evidence::FileRef becomes a SARIF physicalLocation. No new dependencies.
• junit — to_junit_xml emits a Jenkins/Surefire JUnit XML document. Every check becomes a <testcase>; fails get a <failure> child, skips get a <skipped/> child, warns are emitted as passing testcases (JUnit has no native warn representation; use SARIF for warns). No new dependencies.

Verdict rules

Computed by Report::overall_verdict():

Wire format and JSON Schema

The canonical wire format for both Report and MultiReport is JSON. A JSON Schema document (Draft 2020-12) describing every field is shipped in the crate at schema/report.schema.json.

The schema is the contract for cross-language consumers: a TypeScript / Python / Go / jq user can write tooling against a Report without touching Rust. The schema covers all of Report, MultiReport, CheckResult, Verdict, Severity, Evidence, EvidenceData, and FileRef, with field-level descriptions.

CI validates a generated sample against the schema on every run via scripts/validate_schema.py and the schema_sample example.

The dev-* suite

dev-report is the foundation. The other crates produce reports in this schema:

• dev-fixtures - test environments and sample data
• dev-bench - performance measurement and regression detection
• dev-async - async-specific validation
• dev-stress - high-load stress testing
• dev-chaos - failure injection and recovery testing
• dev-tools - umbrella crate with feature gates

Status

v0.9.x is the pre-1.0 stabilization line. The schema is at schema_version = 1 and stays there through this line. Minor additions remain possible before 1.0; the 1.0 release will pin the schema and follow strict semver.

Minimum supported Rust version

1.85 — pinned in Cargo.toml via rust-version and verified by the MSRV job in CI. (Bumped from 1.75 because transitive dependencies in the suite require edition2024, stabilized in Rust 1.85.)

License

Apache-2.0. See LICENSE.